Thank you. So let's set the stage for our conversation. You walk into an exam room, you grab a little hand sanitizer, and you introduce yourself to a patient whose chief complaint is something ambiguous like cough and shortness of breath. What's going through your mind?

There's a couple kinds of questions that are going to go through my mind. As an emergency physician, one main question we usually ask is sick or not sick, right? Very easy question, two possible answers, but critically important, to identify the patients who are clearly ill or to identify the patients who are not clearly ill but at risk of becoming ill, as opposed to those patients who, with some simple supportive care, will do just fine on their own, plus or minus some oral antibiotics, et cetera. So the kinds of questions we ask is, what's making the person sick? Is it an infection? If it's an infection, what kind of infection is it? And then is that, again, is that patient going to make them sick? And what kind of supportive care will they need? Will they need intravenous fluids, intravenous antibiotics, respiratory support that needs to happen in a hospital and maybe even in an ICU setting? Or is it chicken noodle soup plus or minus some home antibiotics and they'll do okay with supportive care at home? And finally, if they are really critically ill, what kind of resuscitation tools do they need? And then something that we'll get a little bit to address a little bit more at the end of the talk, which are, are there any novel or advanced therapies that are going to be helpful for this person? Recognizing that right now, hopefully not in the future, but right now, the number of novel therapies are limited.

So given all of those questions swirling around, how does diagnostic testing help you answer those questions?

I think I propose it could be broken down into three main buckets, so to speak. The first one is I like to call it diagnosis of infection or sepsis, which is, is the person sick from an infection? The second one is prognosis. Just how sick are they or how sick are they going to become? And the third one is treatment and response to treatment, which is what are the treatments that they need? And as we treat the patient, how are they responding to that therapy? In other words, is the current status of treatment okay? Or if they're not responding well to treatment, do we need to then deviate?

So in thinking about lab tests in the context of establishing whether a patient has an infection, there are two complementary approaches. Lab tests that help identify the pathogen itself, and then tests that help measure some aspect of the host response to infection. So in the last few years, you've worked a lot on various host response diagnostics. Can you explain the difference between pathogen identification tests and host response tests, and what drew you to host response diagnostic approaches in particular?

Yeah, so great question. So for pathogen ID, that's going after the bug itself. For example, BioFire PCR looking for bacteria themselves, or in the more simple of cases, COVID PCRs, going after COVID, going after the pathogen itself, what's making the person sick. This is in contrast to the host response, which is how is the body responding to the infection? And this goes more about the pathophysiology of the response. And with that, looking some at is there a different host response or body response to different types of infection? And secondarily, what can we learn about the host response for the prognosis bucket for how sick are they going to become? That's just been my personal interest is more in the host response. To me, at the bedside, it's what we see. And finding ways to uncover a little bit more about how the host is responding at the bedside is part of the interest there. And finally, in some ways, when you get more into sepsis, you have the cause of the infection. But really, it's the host response that is probably the biggest differentiator of how sick somebody is. And so for me, that's always been the area of interest that I'm more focused on.

Great. So once you've decided what test you want to order, how do you then incorporate those results into your patient assessment?

So this is, I mean, this is a little theoretical and academic, but I like the Bayesian approach, which is essentially I'm going to go in and make an assessment of a patient, maybe history and physical exam. They have a cough. They have a fever to 102. They're tachyptych. that's a person sick with an infection or something a little bit more subtle and less compelling. But essentially, I'd like to walk out of the room and say, I have an assessment that let's just oversimplify. My probability for an infection is low, medium, or high. I'm then going to pull in the results of those tests to help inform me and say, okay, what are my next steps going to be? So, for example, if I go in and think someone's sick and the host response confirms that, that's pretty helpful they're going in the wrong direction. If I go in and say, this person looks pretty good, but then my host response markers say, we suspect high probability of infection and or severity, that's where you're going to circle back into the room, double click on the patient, and really try and attenuate which is kind of more informative and how do I put those two pieces of information together? In other words, did I miss something on my exam? Do I need to keep them a little longer to see what direction they're going to go, et cetera? So it's really coming up with a certain assessment, pulling in the results of the testing as an adjunct to help the clinician, and then figuring out how that can get you into the best place.

So, because you mentioned test accuracy as an important part of your clinical reasoning in this paradigm, I'd like to pause here and to review a few key concepts in diagnostic accuracy that I think will help us with our conversation about specific diagnostics. So, getting down to brass tacks, what we want to know about any diagnostic method is how well it matches up with the patient's true disease status. So very simply, when a patient has the disease, we want the test to be positive. And when the patient doesn't have the disease, we want the test to come back negative. And we typically use four ratios to summarize the test performance. We look at what proportion of the test results are positive in patients where the disease is present, which is sensitivity. We look at what proportion of test results are negative when the disease is absent or specificity. We look at the positive test results and see how many of those patients had the disease or positive predictive value. And we look at all the negative test results and see how many of those patients did not have the disease or negative predictive value. But these outputs are all about the test itself, not about correctly diagnosing one patient. So to answer the question, what are the chances that a positive test means that the patient actually has the disease? Or what are the chances that a negative test means that the patient doesn't have the disease? We turn to likelihood ratios. And so the positive likelihood ratio is the probability of a true positive result divided by the probability of a false positive result. And the negative likelihood ratio is the probability of a false negative result over the probability of a true negative result. So how does all this math turn into an actual clinical decision?

This is, you know, likelihood ratios are essentially asking the question, if my test result is positive, how much should it really push my revised probability? In other words, if I think there's a 50-50 chance somebody has an infection and that diagnostic test has a likelihood ratio close to one, no matter what the response is, it shouldn't really move my needle on whether now my revised probability because that particular test in that particular range isn't as helpful. However, if the likelihood ratio is, say, 10, that's really good. And so therefore, a positive test, I should really say, okay, I'm now going to have a much higher suspicion for that person having the outcome of interest. So in general, this is just a rule of thumb. But if the likelihood ratio, so a negative likelihood ratio addresses a negative test. So if it's between 0.5 and 1, it doesn't really alter it that much. the test isn't reliable enough to really alter a tremendous amount. 0.1 to 0.5 means that negative test, I'm really going to start to believe. And then if the likelihood ratio is less than 0.1, then it's really compelling that the negative test means much lower probability of having the disease or the outcome. When we look at the positive likelihood ratios, one to two, it moves it a little bit. Two to 10, now you're really starting to get helpful. And then greater than 10 means highly useful. In other words, I really believe that positive test. And so really you have to integrate these likelihood ratios with your probability of disease to try and put the whole picture together. Now, part of the reason we're bringing it up for this particular talk is if you look at the FDA approvals across the board of the diagnostics, the likelihood ratios are kind of front and center of the way that these results are being reported. And in particular, a lot of the not all, but a lot of the diagnostics for sepsis have gone into reporting the likelihood ratios in multiple bands. So there's sort of like a very low, low, moderate, and high probability with escalating likelihood ratios across the board there. So as we talk about the different tests, we're going to be referring to likelihood ratios because that's the way the tests are being reported. And you can kind of, again, use this as a don't worry too much about the math right now, but you can use this as a rule of thumb, which is 0.5 to 1 is not as helpful. 0.1 to 0.5 for a negative test is really starting to become useful, and then less than 0.1, highly useful, and then the corollary for the likelihood ratio, positive.